Generating Structured Adversarial Attacks Using Frank-Wolfe Method

no code implementations • 15 Feb 2021 • Ehsan Kazemi, Thomas Kerdreux, Liquang Wang

White box adversarial perturbations are generated via iterative optimization algorithms most often by minimizing an adversarial loss on a $\ell_p$ neighborhood of the original image, the so-called distortion set.

Adversarial Robustness